How to implement AI in your business
A 5-step guide on how to build an artificial intelligence solution for your business, and do you really need to?

Things to consider before starting an AI project.
Nowadays, artificial intelligence, or simply AI, has become impossible to ignore. The latest advances in machine learning, computer vision, and chatbots have seen a surge in popularity both in business and consumer applications. These rapid advancements have brought AI to the agenda of business owners which now need to decide whether AI can transform their company and customers' experience and how to make it happen.
    Is it the right time to start with AI?
    Global revenues in the AI market, including software, equipment, and services, will grow by 16.4% year on year in 2021 to 327.5 billion US dollars according to the international, semi-annual artificial intelligence tracker of the International Data Corporation (IDC). With a five-year compound annual growth rate (CAGR) of 17.5%, the market is expected to break the $500 billion mark by 2024 and total an impressive $ 554.3 billion in revenues.
    While the majority of AI technologies don't move beyond the ideation stage yet, few technologies such as machine learning, computer vision, and chatbots are gaining more and more momentum as they are actively tested in real application environments.

    We are at the very beginning of the AI story and becoming an AI pioneer could create immense business opportunities or significant downside risks due to the high investments and over expectations. Is there a middle ground and what is the right AI strategy for business owners?


    Ok, Google! How can I benefit from integrating AI into my applications?

    An easy surf through the web will advise that some business challenges have been solved with the help of AI. You will further find some feedback for successful and failed implementations of AI capabilities:

    • Increase sales
      Benefit from the vast amount of customer data you have collected over the time and use the insights to derive actionable recommendations. These insights ultimately increase your customer retention rate and, therefore, your profits.
    • Improve the quality of customer service
      Process the tons of incoming data and analyze user behavior to segment and customize your service offering as granular as your audience desires it.
    • Identify risks and predict trends
      Improve the way your business handles uncertainty and increase your business forecasting accordingly.
    • Reduce labor costs
      AI-driven process automation helps you to release your employees from routine work and enable them to focus on priority tasks.


    These thoughts will give you the first level of understanding if you need extra innovation in your business operations. But do you think AI will give you a quick move ahead of competitors? If so, then it is the right time to find out about the practical steps on this way.
    Are existing solutions good enough or do you require a custom project?
    People tend to have unrealistic expectations of how AI works. As technology developed by humans can only operate within the boundaries the technology had been taught, the magical AI remains a fantasy. Nonetheless, when applied correctly, AI can create astonishing results by transforming how business processes are designed. Implementing AI, however, is neither a straight-forward exercise nor as exciting as it might sound. It all starts with pretty boring steps, namely:

    1. Explore the business weaknesses and limitations you want to improve
    2. Look around and see how others approach similar problems
    3. Check if regular digital solutions without built-in artificial intelligence for your task already exist
    4. Define what the main goal of the business process is and what the desired gains from investing in an AI application are
    Traditional development techniques capable of solving your task are most likely more cost efficient and quicker implementable than developing a new AI tool. That is why AI solutions should only be chosen when a simpler, straight-forward solution cannot guarantee a satisfactory, long-term solution.
    Can AI truly solve your problem?
    Before developing an AI solution, you should test whether AI can truly solve your problem by creating a so-called baseline model and testing for human-level performance. Here is a step-by-step instruction:

    1. Articulate the problem you plan to resolve with an AI solution and define the accuracy level that would be considered appropriate for the successful completion of your business task.
    2. Prepare a labeled dataset.
      This is a collection of input data you will use for training the initial ML algorithms. What that may be? Let's say, if you want it to recognize flowers on images with AI – pick up 20 images for each type of flowers and label them accordingly; if you plan to identify trends with AI – pick up scenarios and add the attributes. Sometimes you have to gather a dataset manually and sometimes you can find an open-source dataset you need.
    3. Find out what the human-level performance level is or create a baseline model.
      Testing the human-level performance means nothing else than identifying the level of accuracy human beings would achieve when doing the same task that is expected from the future AI solution. In other words, invite 20 random people that are neither familiar with you nor your business and ask them to perform the exact same task that is expected from your future AI solution. This might be the identification of a specific type of flower. If humans fail to achieve the necessary accuracy levels, AI will fail too for sure. The accuracy level you achieve during such tests will show you the actual level of AI performance you will get with your future AI solution.

      For specific types of datasets (e.g. tables) it is impossible to measure human-level performance. In this case, you need to hire an engineer who will create a baseline model to test your dataset using an off-the-shelf neural network already used for similar tasks.
    4. Analyze the level of accuracy of the baseline model or evaluate the human performance achieved.

      If the test results did not meet your expectations, even though you made sure the quality of data samples was good enough (clear pictures, no missing data), it means that AI technology cannot solve the task either. If you are satisfied with the score, it's time to move to the next step and prepare the entire data set.

      The baseline model is also useful to evaluate the quality of the final model and to quickly eliminate errors and problems. If your final model performs worse than your baseline or does not achieve the accuracy level showed by the human level performance, you will need to keep improving it.
    Is your dataset of a high enough quality to be used in an AI application?

    Figure 1. Cone of Uncertainty.
    Data is the most critical aspect when it comes to training AI algorithms. Following the saying "garbage in - garbage out", it is important to control for data quality from the very beginning on. Therefore, it is not surprising that about 80% of all time spent on developing a new AI solution is allocated to data gathering, data cleaning and data labeling.

    Long story short, the more data you can offer, the better the project will succeed. The key to AI is the large amount of data. Research shows that on small datasets, traditional forecasting methodologies such as multiple regressions work similarly well as AI solutions. However, once big data comes into play, a competitive advantage can be achieved by using neural networks and other AI methodologies.


    But where to find data? Combine internal and external data sources. When gathering your data, don't forget the free data sources such as:

    • Open datasets prepared by enthusiasts
    • Google search and data parsing
    • Cooperation with laboratories, or industry organizations that might see potential in partnering with you
    • Web scraping
    • National and international institutions (e.g. governments)


    To prepare a dataset for machine learning, you need to label the data or, in other words, provide the correct context so that the AI model can learn from the data. The data labeling procedures depend on the AI use case and can vary significantly. Let's simplify this through three examples:

    1. If you intend to use AI to distinguish between dogs and cats in photos, you should be ready to gather a large set of images with different background, lighting conditions, angles and for each of them indicate whether it contains a dog or cat (=label them).

    2. If you want to detect the presence of a tumor in MRI scans, evaluate where you can access a rather big number of scans. How many scans you need depends on various questions such as whether you want the algorithm to feature the tumor on the x-ray or not. Then label the scans with a special software to create actionable insights. Through the labeled scans, the algorithm ultimately learns how a tumor looks and can identify them accordingly.
    3. Adding AI-functionality to the scoring procedure of a consumer's creditworthiness would require having the database of borrowers labelled based on whether the debt has been paid off already or the customer is in arrears.

    You may complete these tasks by yourself or hire a Data Labeling Quality Specialist or Data Scientist to ensure the success of your project.
      Ready for the proof of concept?
      Congrats, if you got here! This means that you have a rather complex and promising task for an AI algorithm at hand, have gone through the baseline exercise, and have been gathering a dataset accordingly. Now is the right time to continue with a team of AI experts to create a POC: thoroughly work on testing approaches, models and playing with model parameters to make the best use of your dataset. Further, AI experts are able to estimate the business impact of the new AI solution and evaluate whether it makes economic sense to pursue the project.

      The sad truth about AI is that no one can guarantee success before the proof-of-concept stage is over. Therefore, we ask you to think about the AI output accuracy level you would tolerate, what acceptance criteria should be for the project and what dataset you will use for testing your new model. We know that defining the right level of accuracy can be daunting, that is why we can collaborate with you to combine our AI know-how with your industry expertise and define the right parameters together.

      If the POC is successful, you can gather additional data and understand how to better deploy the model prior to moving the project to the next stage of neural network development.

      At Roonyx, we enjoy AI experiments with our customers. During the POC stage, we evaluate how economic and efficient different approaches and models can be for a customer's task and how feasible the idea is.


      In a nutshell:
      Are you thinking about AI to change your business?

      1. Articulate the business problem and how an AI-solution should solve it; identify the data you will need
      2. Check existing solutions and decide if they are good enough or you require a custom project
      3. Gather relevant data, make a baseline and check human-level performance
      4. Gather and label an extensive dataset for your AI algorithm
      5. Come to AI-experts for Proof-of-concept

      See also